Plasticity genes and plasticity costs: a new approach using an Arabidopsis recombinant inbred population

Authors

  • Hilary S. Callahan,

    Corresponding author
    1. Department of Biological Sciences, 3009 Broadway, Barnard College, Columbia University, 3009 Broadway, New York, NY 10027–6598, USA;
      Author for correspondence: Hilary S. Callahan Tel: +1 212 854 5405 Fax: +1 212 854 1950 Email: hcallahan@barnard.edu
    Search for more papers by this author
  • Natalie Dhanoolal,

    1. Department of Biological Sciences, 3009 Broadway, Barnard College, Columbia University, 3009 Broadway, New York, NY 10027–6598, USA;
    Search for more papers by this author
  • Mark C. Ungerer

    1. Division of Biology, Kansas State University, Manhattan, KS 66506–4901, USA
    Search for more papers by this author

Author for correspondence: Hilary S. Callahan Tel: +1 212 854 5405 Fax: +1 212 854 1950 Email: hcallahan@barnard.edu

Summary

  • • Earlier flowering is triggered by vernalization in some but not all Arabidopsis ecotypes, often reflecting allelic variation at the FRIGIDA (FRI) locus. Using a recombinant inbred (RI) population polymorphic at FRI, we examined fitness consequences of variation for plasticity.
  • • Flowering and fitness were scored for 68 RI genotypes following full and partial vernalization treatments. Within-environment and mixed-model anovas estimated variance components for a genotype effect and a G × E term, respectively. Selection analyses examined whether delayed bolting increases fitness; a plasticity costs analysis asked whether increased plasticity lowers fitness. We also explored whether trait QTL had environment-specific effects, colocated in the immediate vicinity of FRI, or overlapped with fitness QTL.
  • • Selection may favor fri alleles and constitutive early flowering, especially in conditions that only partially vernalize plants. Plasticity costs, detected only after partial vernalization and only marginally significant, were nonetheless consistent with FRI-FLC function.
  • • We discuss how information about QTL with environment-specific effects, fitness QTL, and knowledge about plasticity genes can improve interpretation of selection or plasticity cost analyses.

Introduction

Phenotypic plasticity is the expression of alternative phenotypes in response to varying environments. Developmental and evolutionary ecology studies addressing phenotypic plasticity often address a common set of questions, including whether there is natural genetic variation for plasticity, the raw material for the evolution of plasticity, and whether phenotypic plasticity is indeed an adaptation to divergent selection regimes. This typically involves scoring both plastic traits and environment-specific fitness in arrays of genotypes replicated within and across multiple environments. After estimating selection differentials or gradients, it can be determined whether plasticity results in phenotypes that better match selection regimes (Pigliucci, 2001; Schmitt et al., 2003). These data can also be used to address a third relevant question: Is plasticity costly? (i.e., all else being equal, are more plastic genotypes less fit than less plastic genotypes?) (Van Tienderen, 1991; DeWitt, 1998; Relyea, 2002)? It is well worth asking this latter question given a discrepancy between the theoretical importance of plasticity costs and evidence about their frequency and magnitude (Scheiner & Berrigan, 1998), and given the straightforward methods available to test this (DeWitt et al., 1998).

Studies documenting significant plasticity costs are much less common than those finding costs that are small in magnitude or negligible, suggesting that past selection has minimized costs (DeWitt, 1998; Sultan & Spencer, 2002). Yet studies typically sample natural populations and therefore lack information about genetic machinery underlying plasticity, including so-called ‘plasticity genes’ or loci that directly respond to environmental stimuli, triggering phenotypic plasticity in one or several traits (Pigliucci, 1996). An exceptional study in this literature documented significant plasticity costs by comparing the viability of transgenic strains of Drosophila melanogaster possessing either enhanced or disabled expression of heat shock proteins (Feder, 1999). Based on that study, it has been suggested that plasticity costs may be detectable only when large, single gene effects are the underlying genetic basis for plasticity (Berrigan & Scheiner, 2004). Yet it is certainly possible that genes of large effect were the basis for plasticity in studies involving natural genotypes, including studies that failed to detect costs. Indeed, Agrawal (2001) has discussed the difficulty of interpreting very small or negligible plasticity costs in the absence of information about plasticity machinery. Clearly, more studies conducted with well-characterized genotypes are needed.

Studies involving transgenic organisms may lack biological realism or be subject to numerous challenges (e.g. confounding effects of allelic variation with genetic background or position effects: Tatar, 2000). Here, we report a new approach for quantifying and interpreting plasticity and its associated costs. The research described in this paper examines two highly plastic flowering time traits in a recombinant inbred population of Arabidopsis thaliana, exposing genotypes to manipulated temperatures mimicking periods of overwintering. This allows assessing each genotype's response to vernalization. It uses a mapping population of recombinant inbred lines (RILs) harboring appropriate variation for plasticity. In this genetically well-informed framework, negative results (i.e. failure to detect costs) may be more interpretable compared with studies sampling directly from natural populations. A comparable approach would be feasible for any species for which RI populations can be created.

Our specific study system offers several additional advantages. First, conducting genotypic selection analyses and plasticity cost analyses with Arabidopsis is generally straightforward. Mean trait values and mean lifetime fitness are readily quantifiable in replicated genotypes. Several studies have successfully used natural genotypes or RILs to conduct genotypic selection analyses, including studies focusing on flowering time plasticity induced by shading (e.g. Dorn et al., 2000; Callahan & Pigliucci, 2002), responses to herbivores (e.g. Mauricio & Rausher, 1997; Weinig et al., 2003), and other factors (reviewed by Pigliucci, 2003). To date, these techniques have not been used to rigorously examine the adaptive significance of vernalization requirements, or to probe the phenological or climatic factors favoring its maintenance or loss. This is somewhat surprising since the underlying mechanism of plasticity to vernalizing temperatures has been intensively studied (Henderson et al., 2003) and among-population variation for vernalization requirement has been extensively documented (Karlsson et al., 1993; Nordborg & Bergelson, 1999; Stinchcombe et al., 2004).

Accordingly, a second advantage of using RI populations of A. thaliana is that vernalization responses are known to exhibit variable plasticity among ecotypes. Responses are typically dichotomized as either present (winter annuals) or absent (summer annuals). In winter annuals, flowering is chronologically and developmentally delayed in the absence of vernalization. By contrast, in summer annuals, flowering occurs early regardless of vernalization treatment (Nordborg & Bergelson, 1999). Most of the variation in vernalization-mediated flowering habit is controlled by the FRIGIDA (FRI) locus and its interaction with FLOWERING LOCUS C (FLC). Elucidation of FRI and FLC function has led to extensive characterization of FRI alleles in many wild ecotypes (Johanson et al., 2000). Functional and nonfunctional alleles of the FRI gene segregate in an established population of recombinant inbred lines (RILs) (Wilson et al., 2001); this population harbors extensive, bi-modal variation for flowering habit. The large number and diversity of RILs within a mapping population can increase statistical power for testing the significance of selection gradients, a particularly crucial issue in searches for plasticity costs (DeWitt, 1998; van Kleunen et al., 2000; Pigliucci, 2001).

A third, general advantage of our approach is that RI populations are well-suited to quantitative trait loci (QTL) analysis, including efforts to associate plastic phenotypes and fitness with QTL–environment interactions (Weinig & Schmitt, 2004). Here, in multiple RI lines we score the environment-specific means for a trait, the mean plasticity associated with that trait, and mean fitness. We then examine QTL, whether effects of QTL are environmentally variable, and whether trait or plasticity QTLs colocate in the same chromosomal regions as well-characterized plasticity genes (e.g. FRI or other well-studied environmental signal transduction genes). This makes it possible to look for overlap between chromosomal regions harboring fitness QTL and the QTL affecting plastic traits. The presence of such overlap is a first step in determining whether genes regulating plastic traits exert pleiotropic effects on fitness.

Our study analyzed a single experiment by integrating four quantitative genetic analyses. First, it quantified two flowering traits and their associated plasticity using the Col-gl1 × Kas recombinant inbred lines, originating from parents lacking (Col-gl1) and possessing (Kas) a vernalization requirment (Wilson et al., 2001). Because this population harbors a polymorphism at the FRI locus, RILs possessing a functional FRI allele should respond to brief, partial vernalization by showing delayed flowering, while RILs with a null allele should exhibit constitutive early flowering. A lengthy, full vernalization treatment should minimize this disparity, with all RILs flowering early. Second, after confirming these predictions, we conducted a genotypic selection analysis using environment-specific relative fitness data to estimate and test the significance of selection differentials and gradients on these two flowering traits. Third, by combining this information with data about the plasticity of each RIL, we quantified plasticity costs with multiple regression analyses within each environment. Fourth, using QTL mapping approaches, we examined the genetic architecture of plastic flowering-time traits, confirmed the importance of the FRI locus, and examined potential pleiotropic effects of trait QTL on fitness.

Methods

Plant materials

Wilson et al. (2001) generated a population of 129 Arabidopsis thaliana RILs by crossing two parental ecotypes: Columbia carrying a glabrous 1 allele (Col-gl1) and Kashmir-1 (Kas-1). These ecotypes were initially selected because of differences in pathogen resistance (Wilson et al., 2001), but also differ markedly in their vernalization response. Kas-1 is late-flowering and has a putatively functional FRI allele. Col-gl1 is early flowering and carries a null fri allele with a 16-bp deletion (C. Alonso-Blanco, pers comm.) The Arabidopsis Biological Resource Center (ABRC) provided F6 seed and preliminary screening was used to assess each RIL's flowering habit in three different vernalization treatments (Callahan, unpublished data; more details below). Our growth facilities and experimental design could accommodate a maximum of 72 replicated genotypes, chosen to represent as many vernalization-sensitive RILs as possible and a random sample of RILs lacking a vernalization requirement. Upon request, a complete list of RILs in this study is available from the Correspondence.

Vernalization conditions

For vernalization-sensitive Arabidopsis ecotypes, extending the duration of exposure of cold further accelerates flowering, but only up to a point; the graded response to vernalization eventually ‘saturates’ (reviewed in Michaels & Amasino, 2000). In the cold treatments used in our growth chamber experiments, which were simplified versions of field conditions, we aimed to achieve ‘partial’ or ‘full’ vernalization. These treatments were calibrated in a preliminary screening, mentioned above, that included 16-, 32- and 39-d cold treatments at 5°C. Plastic genotypes showed delayed bolting in the 16-d treatment and a similar acceleration of bolting in the 32- and 39-d treatments. We therefore chose 16 d as the ‘partial’ treatment and 32 d as our ‘full’ vernalization treatment. In both the preliminary screening and in this study, the two treatments served as an appropriate contrast that allowed us to characterize each RIL's plasticity.

Growing conditions

To initiate the life cycle of each individual plant, we used a sterilized paintbrush to plated 16–20 evenly spaced seeds of each RIL onto deep 100-mm Petri plates filled with agar-based (0.8 g l−1) growth medium (full strength MS Basal Salts, Sigma, St Louis, MO, USA). Following imbibition in the dark at 5°C for 2–3 d to promote and synchronize germination, we allowed plants to develop for 12 d at 20°C, at which point plants began initiating true leaves. Plates then were vernalized at 5°C; we used staggered start dates so that replicates in partial (16 d) and full (32 d) vernalization treatments were simultaneously switched from 5°C to 20°C conditions. At this time, all plants were also transplanted into individual soil-filled pots (#2 growth medium, Fafard, Agawam, MA, USA). Individual pots were randomly assigned to positions within standard glasshouse flats containing 72 pots per flat. Replication occurred in eight randomized spatial blocks consisting of two flats each (n = 1152 pots). Blocking aimed to control for lighting variation within our walk-in growth room, lit by four banks of fluorescent light fixtures containing four 40 W GROLUX tubes (Sylvania, Danvers, MA, USA), four flats per bank. Throughout the experiment, a 16 : 8 photoperiod was maintained. After transplanting to soil, all pots were bottom watered approximately twice weekly.

Phenotyping RILs

After transplantation, all pots were checked at least every other day for bolting, which was scored as the appearance of c. 3 mm of elongating inflorescence. At bolting, we recorded two flowering traits: first the number of days since the plant was returned to 20°C, and second the number of rosette leaves. These estimate, respectively, the chronological and developmental timing of the transition to flowering. Six to seven weeks after germination, when all siliques were ripe and many had shed seed, we harvested inflorescences and recorded total fruit number. If fewer than eight seedlings were available for transplanting, or if a seedling did not survive transplanting, data were recorded as missing. If an inflorescence did not develop after 100 d, bolting day and leaf number data were recorded as missing and we recorded zero fruits. Seeds from four RILs failed to germinate, and these RILs were dropped from analyses.

Data for all three traits were log-transformed to improve normality and heteroscedasticity. For statistical rigor, results of anovas and regressions conducted with transformed data are presented, but summary statistics are reported in terms of back-transformed data. QTL analyses were conducted on both raw and transformed data, with identical results. QTL results are presented on untransformed data, in order to make the effect sizes of QTLs more interpretable.

Quantitative genetic analyses

For three traits (leaf number at bolting, days to bolting, and total fruit number), we used SAS PROC MIXED (SAS Institute, 1999) to estimate restricted-maximum likelihood variance components for the random-effects anova models:

Y = µ + RIL + Block + Residual error

where RIL represents the random effect of genotype and each spatial block has one replicate per genotype. Restricted maximum likelihood variance components were estimated and tested using Wald's Z statistic (Littell et al., 1996). We also used SAS PROC MIXED to estimate univariate mixed-model anovas using the model:

Y = µ + Block + RIL + T + RIL × T + Block × T + Residual error

which extends the within-environment model above by including the fixed effect of vernalization treatment, T (i.e. partial or full) and its interaction with RIL and Block. Variance components were estimated for all random effects, including interaction terms. Because there was only one replicate plant per environment per block, a three–way interaction term could not be included. Models were also fit with a Block × RIL term, which was dropped from the final model because the variance component estimated for this term was always zero. We report the results of tests for RIL, RIL × Treatment, Block, and Block × Treatment based on Wald's Z. The vernalization effect was tested with an F-ratio test in which the RIL × Treatment term was the denominator; degrees of freedom were Sattherwaite approximated to adjust for missing data.

Genotypic selection analyses

Trait means for each RIL within each environment were calculated as an estimate of genotypic means. To estimate relative fitness within an environment, the RIL mean for log-transformed fruit number was divided by the within-environment grand mean (Lande & Arnold, 1983; Rausher, 1992). This estimate of relative fitness was regressed on a focal trait to estimate a standardized selection differential. Two traits were separately modeled: days to bolting and leaf number at bolting. Tests were conducted with and without a sequential Bonferroni adjustment for multiple tests (Rice, 1989), following Relyea (2002). We also conducted bivariate analyses that included leaf number at bolting and days to bolting in a single model, although these should be interpreted conservatively given the strong collinearity between these traits. Quadratic regression was conducted to examine potential stabilizing or disruptive selection, but are not presented because quadratic terms were not significant in any of the models evaluated.

Cost of plasticity analyses

Following Van Tienderen (1991) as modified by Dewitt et al. (1998), we tested for a cost of plasticity using a multiple regression model:

W = X + plX

where W is the RIL's mean relative fitness within an environment, X is the RIL's trait mean within that environment and plX is the RIL's mean plasticity for that trait to the contrasting vernalization treatments as estimated by Falconer's (1990) environmental sensitivity score, calculated as:

(E1i − E 2i)/D,

where D is the difference between vernalization treatments of the means of all RILs, and E1i and E 2i are the means of replicate individuals of the ith RIL in the partial and full vernalization treatments.

A negative regression coefficient for the plX term indicates a cost of plasticity, namely a lower fitness of plastic genotypes relative to nonplastic genotypes, all else being equal. A positive coefficient indicates a cost of homeostasis (Dorn et al., 2000). Past studies have argued that such costs, representing maintenance or genetic costs of plasticity, are likely to be small in magnitude and difficult to detect. We therefore analyze the data both with and without a sequential Bonferroni correction to adjust for multiple tests on multiple traits (Scheiner & Berrigan, 1998; Relyea, 2002). We conducted separate univariate analyses on leaf number at bolting and its plasticity, and on days to bolting and its plasticity because of concerns about loss of statistical power as additional terms are added to multiple regression models, and due to collinearity between days to bolting and leaf number.

Mapping and QTL analyses

A linkage map for the Col-gl1 × Kas RI lines was established by Wilson et al. (2001) using 26 markers. It was expanded upon by Wolyn et al. (2004) through the mapping of 29 additional markers. Raw genotypic data for the combined data sets were obtained from (http://nasc.nott.ac.uk/) and a linkage map established with the program mapmaker/exp 3.0 (Lander et al., 1987; Lincoln et al., 1992). Because all markers were previously mapped to chromosomes (Wilson et al., 2001; Wolyn et al., 2004), our analysis was restricted to confirming marker orders and recombination distances. For linkage groups with < 10 markers, all possible orders were compared and the order with highest likelihood retained. For linkage groups with ≥ 10 markers, subsets of nine markers were used to establish a framework order as described above and then the maximum-likelihood position of each remaining marker was determined. The order of markers on every linkage group was confirmed by permuting marker positions within a scrolling window of five consecutive markers and recalculating likelihoods for all alternative orders as implemented by the ripple command of mapmaker/exp 3.0. Recombination frequencies were converted to map distances using the Kosambi mapping function (Kosambi, 1944).

Mapping of quantitative trait loci was conducted using the composite interval mapping [CIM] (Zeng, 1993, 1994), and the multiple-trait composite interval mapping [M-CIM] (Jiang & Zeng, 1995) functions of QTL Cartographer 1.17 (Basten et al., 1994; Basten, 1999). These procedures test sequentially along chromosomes whether intervals flanked by molecular markers contain a QTL while statistically accounting for other QTLs segregating outside the tested interval. Multiple-trait composite interval mapping allows for the characterization of QTL–by–environment interactions through a joint analysis of measurements of the same trait in different environments (Jiang & Zeng, 1995; Borevitz et al., 2002; Ungerer et al., 2003). We conducted M-CIM for both leaf number at bolting and days to bolting because of our interest in the genetic basis of these two very plastic and strongly correlated traits. Simple within-environment CIM analyses were conducted for fruit number, to match within-environment analyses to examine the adaptive plasticity hypothesis, and to search for plasticity costs.

Genetic cofactors (markers linked to additional segregating QTLs) were selected before mapping by forward selection, backward elimination stepwise regression. A maximum of five cofactors were used in mapping of single traits and a maximum of 10 cofactors were used in analyses of QTL–by–environment interactions. All mapping was conducted using a 10-cm scan window and a walking speed of 1 cm. Experiment-wise significance thresholds for QTL identification were determined by permutation analysis (Churchill & Doerge, 1994; Doerge & Churchill, 1996); one thousand permutations were performed for each analysis.

Results

Genetic variation for bolting time and plasticity

Bolting occurred as early as 4 d and as late as 77 d following the vernalization treatments, and the greatest delay in bolting occurring after the partial vernalization treatment. As expected, bolting occurred on average c. 5 d earlier after the full vernalization treatment (back-transformed means: full, 10.2 d; partial, 15.2; significant fixed effect of vernalization treatment from mixed model anova: F1,66 = 25.7, P < 0.0001). Plants bolted with as few as four and as many as 25 leaves, with the greatest number of leaves associated with partial vernalization. On average, leaf number differed only slightly between treatments (full, 10.4 leaves; partial, 10.2; F1,65 = 0.44, P > 0.5). The number of mature fruits produced per plant was slightly higher under the full vernalization treatment (18.6 fruits) than under the partial vernalization treatment (16.5 fruits) (F1,66 = 5.17, P < 0.05, from mixed model anova with log-transformed data).

In the partial and full vernalization environments, respectively, among-RIL variance accounted for approximately 75% and 60% of total variance for days to bolting, and c. 66% and 50% of total variance for leaf number at bolting. For these two traits, among-block variance was nonsignificant. For fruit number, our fitness trait, the variance component for the effect of RIL accounted for a smaller but still highly significant proportion of total variance (partial, c. 50%; full, c. 10%; Table 1).

Table 1.  Restricted-maximum likelihood estimates of within-environment variance components for the effect of RIL and Block on two flowering traits and fruit number
 VRILVBlockVError
  • * and **

    , significance at P < 0.01 and P < 0.001, respectively.

Partial vernalization
 Days to bolting0.093**0.00170.028
 Leaf number at bolting0.029**00.015
 Number of fruits0.070**0.0510.026
Full vernalization
 Days to bolting0.025**0.00100.015
 Leaf number at bolting0.007**0.000150.0072
 Number of fruits0.025**0.0830.13

For both flowering traits, RILs differed in their plastic responses to vernalization treatments, as demonstrated by a significant variance components estimated for both the main effect of RIL and the RIL × Treatment interaction term (Table 2). For fruit number, there were also significant variance components for the effect of RIL and for the RIL × Treatment interaction, but these terms accounted for a much smaller proportion of overall variance. Fruit number was the only trait for which a significant Block effect was encountered.

Table 2.  Restricted-maximum likelihood estimates of across -environment variance components for the effect of RIL, RIL × Treatment, Block, and Block × Treatment on two flowering traits and fruit number
 VRILVRIL×TreatmentVBlockVBlock×TreatmentVError
  • * and **

    , significance at P < 0.01 and P < 0.001, respectively.

Days to bolting0.036**0.023**0.001400.021
Leaf number at bolting0.012**0.00061**0.0001200.011
Fruit number0.021*0.026*0.062*0.00530.19

Figure 1 depicts the mean reaction norm for both bolting-time traits and reaction norms for eight genotypes with the four highest and lowest environmental sensitivity scores for days to bolting. Genotypes showing delayed bolting in response to partial vernalization treatment generally also showed a developmental delay (i.e. greater leaf number), but in some genotypes there was no delay in days but an increase in leaf number. Negative environmental sensitivity scores indicating a reverse pattern of plasticity occurred in c. 20 RILs, but were very were low (< −1.0). These genotypes, as well as c. 13 RILs that had low positive scores, are essentially nonplastic to vernalization.

Figure 1.

Reaction norms are shown for the eight RILs with the four greatest and four least plastic phenotypes for days to bolting. Mean reaction norms (n = 68 RILs) for days to bolting and leaf number at bolting are indicated by the grey bar. Genotype numbers refer to the last three digits of stock center catalogue numbers (84871–84999).

Fitness consequences of bolting habit

Table 3 summarizes the result of selection analyses. In both the partial and the full vernalization treatments, earlier flowering was favored, as indicated by negative selection differentials that were marginally or modestly significant, respectively. Selection gradients estimated from a bivariate model including both traits also indicated that flowering in fewer days but with more leaves was favored. Gradients were somewhat larger in magnitude in the partial vernalization treatment, but this result was qualitatively similar in the two treatments. (In an ancova analysis, a treatment × bolting-day interaction term was nonsignificant; F1,133 = 0.54; P = 0.46).

Table 3.  Results of selection analyses regressing an estimate of relative fitness against either or both bolting time traits to estimate standardized regression coefficients, interpreted as selection differentials or gradients, respectively
 Leaf number at boltingDays to bolting
  • a

    , significant after Bonferroni correction;

  • b

    b , significant before Bonferroni correction;

  • c

    , marginally significant, P = 0.065 before Bonferroni correction.

Partial vernalization
Selection differential−0.09−0.23c
Selection gradient 0.69ab−0.86ab
Full vernalization
Selection differential 0.01−0.25b
Selection gradient 0.51ab−0.65ab

We did not detect significant selection differentials on leaf number in either environment. In the bivariate analyses in both environments, however, significant positive gradients for leaf number indicated that, for plants bolting on any given day, selection favors those that bolt with more leaves.

Detection of plasticity costs

In the partial vernalization treatment, we detected a positive gradient on the plasticity term (plX) for leaf number and a negative gradient on the plasticity term for days to bolting, but only the latter was significant, and only modestly so (Table 4). In other words, all else being equal, genotypes with greater plasticity for days to bolting had somewhat reduced fitness, but only in the partial environment. Since days to bolting was a plastic trait overall, this negative gradient is interpreted as a cost of plasticity. In the full vernalization treatment, the signs of the gradients on plasticity terms for these two traits were reversed and the magnitude of standardized coefficients were smaller. In both instances, however, gradients were nonsignificant, even before a sequential Bonferroni correction (Table 4).

Table 4.  Costs of plasticity, as estimated by standardized regression coefficients of a genotype's mean relative fitness on its plasticity, from a within-environment multiple regression analysis that also included the genotype's trait mean
 Leaf number at boltingDays to bolting
  • *

    , significant before but not after Bonferroni correction

  • Models were fit separately for the two traits due to strong collinearity.

Selection gradient on plasticity term
Partial vernalization 0.25−0.49*
Full vernalization−0.14 0.12

QTL mapping of plastic flowering-time traits and fitness

M-CIM analysis detected a flowering time QTL with an environmentally variable effect at the top of chromosome 4; this QTL affected both days to bolting and leaf number at bolting (bottom six panels of Fig. 2). The effect of a QTL allele substitution at this chromosomal region, which includes the locus of FRI, was to delay bolting by 6.7 or 2.6 d and to increase leaf number by 2.7 or 1.7 leaves, in the partial and full vernalization environments, respectively. A second, adjacent, QTL was also detected on chromosome 4, but only for leaf number at bolting; the effect of this QTL was also environmentally variable, increasing leaf number by 2.0 and 1.0 leaf after partial and full vernalization treatments, respectively.

Figure 2.

QTL likelihood plots of Chromosomes II, III and IV. For days to bolting (bottom row) and leaf number at bolting (second row from bottom), multiple-trait CIM was conducted and LOD scores (y-axis) are plotted with respect to map position in centimorgans (x-axis). A solid line from the M-CIM analyses indicates whether a QTL was detected in at least one of the two vernalization treatments. A dashed line indicates whether the additive effect of declared QTLs differ significantly across environments (i.e. whether there is QTL × environment interaction). Solid and dashed horizontal lines show corresponding significance thresholds for the two separate tests. Single-trait CIM analyses were conducted on fruit number (fitness) independently in the full and partial vernalization treatments (top two rows), with the solid line plotting the LOD score and a solid horizontal line indicating the threshold for significance. Horizontal black bars above peaks show 2-LOD support intervals.

Fitness QTL were mapped separately for each vernalization treatment; results are summarized in the top six panels of Fig. 2. The 2-LOD support interval for the QTL on chromosome 4, detected only in the full vernalization environment, does not overlap with the QTL detected for flowering traits, indicating that QTLs for flowering time traits and fitness are nonoverlapping. We also detected two nearly significant QTL on chromosomes 2 and 3. Both of these QTLs were detected in a single environment only.

Discussion

Studies documenting among-population variation for a vernalization requirement (e.g. Karlsson et al., 1993; Lee et al., 1993; Nordborg & Bergelson, 1999) have provided valuable information leading to current understanding of FRI-FLC function and its molecular population genetics (Johanson et al., 2000; Le Corre et al., 2002; Michaels et al., 2003; Caicedo et al., 2004; Stinchcombe et al., 2004). Our study corroborates and further integrates these previous efforts in several ways. First, we found the expected variation among RILs for flowering habit. Second, genotypic selection analyses indicated that earlier flowering may be favored, especially when mild overwintering conditions fail to fully vernalize vegetative rosettes. Third, plasticity cost analyses suggest more plastic genotypes may have lower fitness after experiencing mild overwintering conditions, over and above the reduction in fitness that results from delayed flowering. Fourth, mapping techniques detected QTL with environment-specific effects on flowering, including a major QTL that colocated in the immediate vicinity of FRI but that did not overlap with the significant and nearly significant fitness QTL detected. This study has increased our awareness of both advantages and pitfalls associated with developmental and evolutionary ecology studies involving RI populations. As we discuss interpretation of our results in greater detail, we also offer suggestions for refining future analyses and for designing either lab or field experiments.

Adaptive significance of a vernalization requirement

Previous studies have quantified selection on flowering time in Arabidopsis thaliana, and results have varied. In many different glasshouse environments, Clauss & Aarssen (1994) documented positive correlations between vegetative mass and fecundity. A glasshouse study by Dorn et al. (2000) documented selection in high and low density treatments and in the presence and absence of simulated shade. Chronologically later flowering was selected in all environments, often in combination with selection favoring flowering at a later developmental stage (more leaves), at larger size (larger rosette leaf length), or both. Contrasting examples include field and potted-plant studies conducted in shaded and unshaded habitats that found similar patterns of selection, with chronologically earlier bolting consistently favored, sometimes in combination with selection favoring developmentally later bolting (Callahan & Pigliucci, 2002). In a growth chamber study of time to bolting, leaf number at bolting, and size at reproduction in the Col × LerRI population of Arabidopsis, selection gradients favor bolting earlier chronologically but developmentally later and at a larger size (Mitchell-Olds, 1996). The results of these latter two studies are consistent with the one reported here, as well as with life-history theory that predicts the inability of selection to simultaneously increase size and reduce age at first reproduction (Stearns, 1992).

We expected to find genotypic variation, since any given RI line possessed either a functional FRI allele or a fri mutant. In the partial vernalization treatment, among-RIL variation in both flowering time and fitness were magnified, and RILs that inherited a functional FRI allele from the Kashmir parent tended to exhibit delayed flowering. If such a treatment corresponds well with milder or briefer winters in the field, then a fitness disadvantage may be experienced by plastic genotypes (i.e. possessing a functional FRI allele results in plastically and maladaptively delaying flowering). Interestingly, a weak tendency for functional FRI to be less common at low latitudes has been documented, as well as ample evidence for convergent evolution to lose a requirement via mutations in either FRI or FLC (Johanson et al., 2000; Le Corre et al., 2002; Michaels et al., 2003), although at any given latitude there are examples of populations possessing a vernalization requirement.

Whether selection regimes, flowering behavior and fitness estimates associated with full and partial vernalization treatments in the lab reasonably reflect field conditions remains an important and open question. We note that we may have lowered variance for flowering time and fitness by growing plants in individual pots arranged at uniform densities, protected from herbivores and pathogens, amply watered, and not exposed to viability selection (e.g. Pigliucci & Marlow, 2001). Low variance would likely bias against finding significant gradients, plasticity costs, or fitness QTLs. We recently initiated an overwintering field experiment examining vernalization-mediated plasticity in a larger set of Col-gl1 × Kas RILs. It will allow us to examine whether significant genotypic variance components can be detected under field conditions, where environmental variance can be large. It will also investigate whether the mean relative fitness estimates for each RIL in lab studies correspond with mean relative fitness estimated in the field.

Is a vernalization requirement costly?

Previous studies have attempted to detect plasticity costs by either sampling many highly diverse genotypes (Dorn et al., 2000; van Kleunen et al., 2000; Steinger et al., 2003) or by engineering genotypes with well-characterized plasticity machinery (Krebs & Feder, 1997; Feder, 1999). Our study combined these approaches via analysis of genetically well-characterized RI lines. The plasticity cost detected was environment-specific (sensu Sultan & Spencer, 2002), detected only after partial vernalization. Also, the cost detected was not significant after applying a more stringent rejection criteria (Scheiner & Berrigan, 1998; Relyea, 2002). Negative results are frequent in studies attempting to detect plasticity costs, and can be difficult to interpret (Agrawal, 2001). Here, the plasticity costs detected were neither highly significant nor global (sensu Sultan & Spencer, 2002) yet we know that a gene of large effect (e.g. FRI) is the basis for plasticity. Such a result is consistent with past selection minimizing such costs in the population from which parental genotypes were sampled (Dewitt, 1998; Sultan & Spencer, 2002).

If a less stringent rejection criterion is applied, we conclude that the detected costs are significant. Indeed, the cost detected is rather large in magnitude (a gradient of −0.49). Moreover, its environmental specificity (in the partial vernalization treatment only) is consistent with current understanding of the FRI-FLC pathway. Specifically, up-regulation of FLC by FRI could lead to two types of nonmutually exclusive costs (DeWitt et al., 1998): genetic costs (i.e. due to pleiotropic effects of one or both genes) or maintenance costs (i.e. due to energy required for transcription and expression of FLC and downstream machinery, which must be maintained in growing plant tissues after vernalization: Michaels & Amasino, 2000, 2001). Both types are likely to differ between plastic and nonplastic genotypes. Both should also be lower after exposure to cold, which blocks up-regulation of FLC by FRI (Michaels & Amasino, 2000).

If FRI-mediated plasticity results in maladaptively late-flowering phenotypes after milder or briefer winters and if, additionally, early flowering is selected in such environments, then why do FRI alleles persist, particularly at low latitudes? Le Corre et al. (2002) speculate that purifying selection may have acted during the last glacial periods, but since then diversifying selection has targeted the FRI-FLC mechanism. Also, functional FRI alleles may increase tolerance to drought (Stinchcombe et al., 2004), and several lines of evidence indicate pleiotropic effects of FRI and FLC on water use efficiency (McKay et al., 2003). Thus, selection against FRI-mediated plasticity in milder temperature regimes may be counterbalanced by selection to enhance drought tolerance.

QTL analyses

A novel benefit of using RILs for selection gradient analyses and plasticity cost analyses is that QTL mapping techniques can be employed to examine whether QTLs for the traits under study map to the vicinity of well-characterized plasticity genes, whether those QTLs exhibit environment specific expression, and whether they colocated with QTLs for fitness.

We were successful in detecting one QTL that affected both flowering traits and mapped to the immediate vicinity of FRI. Among-RIL variance for fitness apparently mapped to unique QTL rather than overlapping with QTL for plastic traits. This suggests that selection favoring earlier bolting and against plastic RILs was not due to pleiotropy (or tight linkage) of genes affecting both bolting behavior and fruit production. Possibly, days to bolting was correlated with an unscored trait, and a QTL for that trait overlaps with the fitness QTL detected. Alternatively, the small number of lines used in this study could have resulted in insufficient statistical power to detect all pertinent trait and fitness QTLs.

These are examples of how QTL analyses can go beyond verifying that among-RIL variance maps to QTL harboring candidate genes, asking important questions about the adaptive significance of plasticity, plasticity costs, and pleiotropic impacts on fitness exerted by QTL (and candidate plasticity genes) involved in regulating plastic phenotypes. The answers provided by this study are provisional, and detecting multiple QTL, including more fitness QTL and trait QTL with minor effects, is a critical goal for future studies. This will necessarily involve phenotyping a larger number of RILs, possibly from a larger RI population harboring segregating variation at both FRI and FLC (Bay-0 × Sha: Loudet et al., 2002). Ongoing field studies will permit us to examine not only whether there is overlap between trait and fitness QTL in the field, but also whether the same QTL are acting in lab and field conditions (Weinig & Schmitt, 2004). In lab or field studies, working with a larger set of RILs will be desirable to improve not only QTL detection, but also estimation and testing of selection gradients, including gradients used to quantify plasticity costs.

Acknowledgements

Karin Isaacson, Sasha Mazza, Jamie Wesker, Gillian Wolfe and Karen Zimmerman assisted with setting up and maintaining the experiment and with data collection. HC received financial support from the National Science Foundation (IBN 0344518) and a travel and training award from the Molecular and Organismal Research in Plant History Research Coordination Network. Constructive comments on earlier drafts of the manuscript were provided by Sarah Israel, Katrina del Fierro, Hina Zafar, Nile Kurashige, Sonia Sultan and two anonymous referees.

Ancillary